Our patent pending technology arises from reverse-engineering specific subsets of the human visual system and modelling the individual and group responses of neurons to visual stimuli, providing a powerful representation upon which we have built robust inference engines

Our technology uses complex adaptive wavelet transforms overlaid on parallel nVidia Graphical Processing Units (GPUs) to generate real-time cortical key-point descriptors of visual structure that are adaptable, compact and robust. These highly paralleled proprietary (patent pending) algorithms are adaptable to local stimulus patterns in real time, and employ a variety of customised wavelet constructions.

The wavelet transforms model the spatial computation performed by biological neurons in the primary visual cortex of the vertebrate brain on a large scale. We use this neuronal representation to find key interest points in the image that are tied to perceptually relevant visual patterns. Our key-point descriptors capture further imaging invariants and from this pipeline, we build a lookup representation that is both efficient and highly scalable to databases of millions of images.

This provides state-of-the-art visual pattern recognition capabilities which are well-matched to human perceptual abilities.

Our matching technology includes a unique paralleled probabilistic computation, enabling our systems to learn.